Evidence for greater oxygen decline rates in the coastal ocean than in the open ocean

Evidence for greater oxygen decline rates in the coastal ocean than in the open ocean D. Gilbert, N. N. Rabalais, R. J. Diaz, and J. Zhang Institut Maurice-Lamontagne, Pêches et Océans Canada, 850 Route de la mer, Mont-Joli, G5H 3Z4, Québec, Canada Louisiana Universities Marine Consortium, 8124 Highway 56, Chauvin, LA 70344, USA Virginia Institute of Marine Science, College of William and Mary, Gloucester Point, VA 23062, USA State Key Laboratory of Estuarine and Coastal Research, East China Normal University, 3663 Zhongshan Road North, Putuo District, Shanghai 200062, China Received: 1 September 2009 – Accepted: 2 September 2009 – Published: 18 September 2009 Correspondence to: D. Gilbert (denis.gilbert@dfo-mpo.gc.ca) Published by Copernicus Publications on behalf of the European Geosciences Union.


Introduction
Several studies suggest that oxygen levels are generally decreasing both in the coastal ocean (Diaz and Rosenberg, 2008) and in the deep ocean (Keeling and Garcia, 2002).This is cause for concern as lower concentrations of oxygen have adverse effects on marine life, ranging from reduced growth and reproductive capacity to habitat avoidance and ultimately death.At some coastal sites, eutrophication (Nixon, 1995) from nearby rivers appears to be the main cause of hypoxia (Kemp et al., 2005;Rabalais et al., 2002).But at other sites on the continental shelf and in the deep ocean, changes in ocean circulation (Gilbert et al., 2005;Monteiro et al., 2006) or in winter ventilation (Whitney et al., 2007)  To make things worse, biogeochemistry models embedded within ocean global circulation models (OGCM) generally predict that oxygen concentrations in the ocean will decrease in the coming decades as a consequence of global warming (Fr ölisher et al., 2009).There are early indications that this may already be detectable (Stramma et al., 2008;Johnson and Gruber, 2007), but interdecadal changes in ocean circulation could have played a role in these trends (Fr ölisher et al., 2009).
According to Diaz and Rosenberg (2008), the number of hypoxic sites around the world reported in the scientific literature has increased about an order of magnitude (40 to 400 sites) from the 1960s to the present.This seems to indicate that hypoxia is becoming much more widespread and more severe relative to the historical literature.
But we need to be careful in order to interpret this finding correctly.The fact that more papers dealing with coastal hypoxia started to appear since the 1990s could also indicate that scientists have devoted more efforts to detecting and reporting hypoxia.Do the new reports of hypoxic conditions reflect a truly widespread phenomenon, or do they reflect a greater tendency for scientists to report low oxygen conditions, because of the associated adverse ecological implications, than to report stable or increasing oxygen levels?
In this paper, we compile global oxygen trend statistics from a coastal band (0 to 30 km from shoreline), from a transition band (30 to 100 km from shoreline) and from the open ocean (>100 km from shoreline).In each distance category, the mean trends, median trends and percentage of negative trends are calculated from oxygen time series published in refereed journals, and from oxygen time series obtained from a global oxygen database.In Sect.2, we provide the details of our search for papers containing plots of oxygen time series spanning at least one decade, and how we determined trends from these time series.We also describe the methods used to merge various public databases in order to construct oxygen timeseries at fixed stations and compute trends from them.In Sect.3, we present maps, a histogram, and statistics of oxygen trends based on the scientific literature, and based on our analyses from the raw data at fixed stations in the global ocean.In Sect.4, we discuss the implications of our findings, and we present the main conclusions in Sect. 5.

Published oxygen time series
Because decadal oxygen variability is important in most marine systems (Garcia et al., 2005), we required that oxygen time series have a duration of at least 10 years.This selection criterion allows us to remove some of the unwanted noise associated with interannual to decadal variability.Being concerned with the large diurnal oxygen cycles that are characteristic of shallow estuaries (e.g., Tyler et al., 2009), we also limited our attention to sites where the bottom was at least 10 m deep, because timing of sampling will influence the oxygen reading and complicate the analyses in these shallower systems.Finally, we restricted our attention to oxygen time series consisting of chemical titration methods only (e.g.Winkler, 1888;Carpenter, 1965;Grasshoff et al., 1999), thus excluding data from electronic oxygen sensors from our analyses.We realize this severely hampers data availability, as more and more people use polarographic or optical sensors to measure oxygen.These oxygen probes can provide excellent data 1985 and later.Some time series were eliminated from our analyses because they did not contain precise depth information (e.g.Lysiak-Pastuszak et al., 2004) for their measurements.In other cases, the individual data points were so cluttered that digitization was virtually impossible to do in a reliable way.One such cluttered time series, from Justić et al. (1987), was nevertheless included in our study by using their trend estimate for the bottom waters of the northern Adriatic Sea and converting the original concentration units of cm 3 dm −3 to µmol L −1 (Table 1).Other oxygen time series with original concentration units other than µmol L −1 were also converted using the factors given in Table 1.For refereed journal papers in which near-bottom oxygen values were plotted, it was sometimes necessary to make an educated guess about the depth of the time series based on local bathymetry (Conley et al., 2007;Peng et al., 2009).In other cases, we had to estimate the climatological depth of isopycnal surfaces (Kang et al., 2004;Konovalov and Murray, 2001;Nakanowatari et al., 2007).In one instance (Lin et al., 2005), we had to correct the originally published oxygen time series (with concentrations over 600 µmol L −1 ) by dividing the published concentrations by a factor of two.The origin of this error in data reporting remains unclear, but one possibility is that the authors reported atomic (O) rather than molecular (O2) DO concentrations.
Figure 1 shows the locations of the published oxygen time series used in our study.The regions of the Kattegat, Baltic Sea and Black Sea represent our most abundant sources of published oxygen data.Other areas with reasonable geographic coverage include the East/Japan Sea and coastal waters around Japan and North America.Unfortunately, a single time series from the southern hemisphere satisfied our selection criteria (Monteiro et al., 2008).A subsample of digitized oxygen time series is plotted on Fig. 2 Since a significant proportion of the above oxygen data can be present in more than one database, we had to eliminate all duplicate records based on metadata information on the latitude, longitude, date and time (UTC) of the measurements.
Based on prior knowledge of long-term ocean monitoring programs, and on visual inspection of plots of oxygen data distribution in space and time, we selected 2132 fixed stations (Fig. 3) for the calculation of oxygen time series from raw oxygen data.For each "fixed" station, we determined the latitude and longitude coordinates of the statistics for the 0 to 300 m depth range.
For the calculation of oxygen trends in a standard reference period (e.g. 1951-2000), we imposed a criterion of 30% overall data availability, and requested at least one year with data in both halves of the time series.For example, for 25-year reference periods, we required at least 8 years with oxygen data.In addition, for the 1951-1975 period, we required at least one year with data before and after the middle year of 1973.Similarly, for the 1976-2000 period, we required at least one year with data before and after the middle year of 1988.At some stations, oxygen concentrations can be very close to or equal to zero at certain depths over most of the 25-year period under consideration.In such cases, unless there is accompanying H 2 S data that can be converted to negative oxygen concentrations, the oxygen trend could be meaningless.There exist some locations where H 2 S data have been collected over many years (e.g.Fonselius and Valderrama, 2003;Konovalov and Murray, 2001), but this is not always the case.To ensure uniformity in our results, we therefore decided to eliminate time series where the median oxygen concentration was less than 10 µmol L −1 .Locations affected by this decision include the deep basins of the Baltic Sea, central Black Sea, Saanich Inlet, Cariaco Basin, and stations located in oxygen minimum zones off Peru, Chile, and India (Fig. 4).

Grouping of trends based on distance from the coast
Due to the effect of Earth's rotation, river plumes generally veer to the right (left) of river mouths in the Northern (Southern) Hemisphere, and then tend to flow parallel to the coastline within about one internal Rossby radius of deformation (Gill, 1982).We take 30 km as a representative value of one internal Rossby radius of deformation between 20 • and 60 • of latitude (Chelton et al., 1998).Based on ocean physics, we expect that most of the impacts of human-induced eutrophication should be felt within this coastal band of 30 km width; albeit, there are exceptions such as the Gulf of Mexico hypoxic zone associated with the Mississippi River plume that extends farther offshore (Rabalais et al., 2007).

Back Close Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Of course, the internal Rossby radius of deformation varies with seasonal stratification strength, but it depends mostly on latitude, with typical values of roughly 10 km at 60 • , 20 km at 45 • , 40 km at 30 • and 80 km at 15 • (Chelton et al., 1998).Given these values, we arbitrarily take 100 km as a representative distance beyond which a parcel of water is no longer under the direct influence of nutrient-rich, buoyant river plumes and coastal currents.We shall refer to this distance category ( >100 km from the coast) as the open ocean.Finally, between the coastal band (0-30 km) and the open ocean, we define a transition zone with distances from the shoreline between 30 km and 100 km.We evaluated distance from the coast for each station by using GSHHS (Global Selfconsistent, Hierarchical, High-resolution Shoreline) shapefiles version 1.3 (Wessel and Smith, 1996), together with the functions distance m and deg2 km m from the Matlab mapping toolbox version 2.7.2.The GSHHS shoreline exists in five spatial resolutions (crude, low, intermediate, high and full).We used the intermediate spatial resolution, which allowed us to determine distance from a fixed station to the shoreline with an estimated accuracy of about 1 km at a reasonable computing cost.

Published time series
Table 2 gives the linear trend of each published oxygen timeseries, together with its latitude, longitude, depth, time period over which the trend was estimated, distance from the coast, and journal reference.An overall summary of these trends based on distance from the coast is given in Table 3.In the 0-30 km coastal band, we find a median oxygen trend of −0.gories of distance from the coast at once, the mean of all published oxygen trends (−0.36±0.52 µmol L −1 yr −1 ) is negative, although not significantly different from zero (Table 3, Fig. 5) at the 95% confidence level.The overall median published oxygen trend is −0.62 µmol L −1 yr −1 , with 73% of all published oxygen time series displaying negative trends.A global map of oxygen trends determined from published time series is shown in Fig. 6.To interpret this map correctly, it is necessary to consider that at certain fixed stations (e.g.station P in Whitney et al., 2007), we have trend estimates from more than one depth level and these are overlain on top of each other.This map shows the general prevalence of negative published oxygen trends, but also highlights a great deal of variability in trend estimates.

Calculated time series
In the 0-30 km coastal band, the mean oxygen trend is negative (at the 95% confidence level) at seven of the ten standard depths, but is not statistically different from zero at the three remaining depths (Table 4).The percentage of oxygen time series with negative trends is greater than 50% at eight of the 10 depths, and the median trend is negative at nine of the ten depths.The number of time series from stations located near small islands are indicated in Table 4 because we do not expect these oxygen time series to be as much influenced by human activities compared with stations located near continents.Impacts of dissolved and particulate nutrients on coastal ecosystems depend on how efficiently these nutrients are retained in nearshore areas.
And clearly, the degree of retainment will be higher in coastal embayments than around a small island subjected to time varying open ocean currents and eddies.The mean oxygen trend over the 0 to 300 m depth range is −0.35 µmol L −1 yr −1 (between −0.47 and −0.24 µmol L −1 yr −1 at the 95% confidence level).
In the 30-100 km transition band between the coastal zone and the open ocean, the mean oxygen trend is negative at six of the ten standard depths, is not statistically different from zero at three depths, and is positive at 75 m depth ( but 75 m.Likewise, the median trend is negative at all depths but 75 m.The mean oxygen trend over the 0 to 300 m depth range is -0.19 µmol L −1 yr −1 (between −0.27 and −0.10 µmol L −1 yr −1 at the 95% confidence level).
Finally, for open ocean stations located at distances greater than 100 km from the shoreline, the mean oxygen trend is negative at four of the ten standard depths, but is not statistically different from zero at the six other depths (Table 6).The percentage of oxygen time series with negative trends is greater than 50% at six of the ten depths, and the median trend is negative at the same six depths out of ten.The mean oxygen trend over the 0 to 300 m depth range is −0.09 µmol L −1 yr −1 (between −0.15 and −0.03 µmol L −1 yr −1 at the 95% confidence level).
Regardless of the distance from coast category, the coefficient of variation (standard deviation/mean) is generally much larger than one.Thus, many time series are required in order to estimate the overall mean trend to within a certain level of accuracy (Hakanson and Stenstrom-Khalili, 2009).
Oxygen trend estimates display greater variability over short reference periods (e.g. 1991-2000) than over longer reference periods (e.g. 1951-2000), as can be seen from the colorbar scales of trends for waters surrounding Japan (Fig. 7) and Europe (Fig. 8).Comparing oxygen trend estimates from the 25-year periods 1951-1975and 1976-2000 in Figs. in Figs.7 and 8, we see qualitative evidence for a greater proportion of negative trends in the 1976-2000 period than in the 1951-1975 period, which we tested statistically.Changes in the estimated trends in DO, either positive or negative, were compared between the two 25-year periods (1951-1975 and 1976-2000) using logistic regression.Data were also categorized into depth intervals (lower limits of each category being 0, 10,20,50,75,100,150,200,250, 300 m) and distance from the coastline (<30, 30-100, and >100 km).There was no multicolinearity between depth and distance from coastline (variance inflation was 1.00 for both variables).This was assessed using a weighted least squares regression of the predicted probabilities from the logistic regression.These statistical analyses were performed using SAS.Overall, increasing depth lowered the odds of a negative trend as did increasing distance

BGD Introduction Conclusions References
Tables Figures

Back Close
Full  1951-1975 and 1976-2000 the odds of a negative DO trend occurring increased by over a factor of 2 (p<0.001).Collectively, these findings are consistent with greater prevalence of coastal hypoxia in more recent years.

Discussion
The median published trends given in Table 3 for the 0-30 km band (−0.98 µmol L −1 yr −1 ), the 30-100 km band (−0.88 µmol L −1 yr −1 ), and the 100 km+ band (−0.54 µmol L −1 yr −1 ), are more negative than the median trends for any single depth in Tables 4, 5 and 6 respectively.This suggests that in general, scientists have a greater tendency to publish papers containing oxygen time series plots when the trend is strongly negative than when the trend is either neutral or positive.This tendency to preferentially report negative oxygen trends appears to be more pronounced in the open ocean than in the coastal ocean.Indeed, for the coastal ocean, the percentage of published negative oxygen trends (70.7%, between 54.5% and 83.9% at the 95% confidence level), is very similar to the percentages of negative trends computed from the raw data of the global oxygen database (Tables 3 and 4).On the other hand, in the open ocean, the percentage of published negative oxygen trends (77.5%, between 61.5% and 89.2% at the 95% confidence level), is higher than nine of the ten percentages of negative trends computed from the raw data of the global oxygen database (Tables 3 and 6).
A number of biogeochemical models embedded within ocean circulation models (Sarmiento et al., 1998;Matear et al., 2000;Bopp et al., 2002) predicted that as a consequence of global warming, oxygen levels in the ocean will decrease at a rate that is typically three to four times faster than one might expect based on changes in oxygen solubility alone (Garcia and Gordon, 1992).For instance, Fr ölisher et al.These modelled trend values are not inconsistent with the mean trend value of −0.09±0.06µmol L −1 yr −1 obtained in this study for the open ocean (Table 6).However, detailed comparisons between our results and biogeochemical model predictions cannot be made with confidence due to the poor spatial coverage of the global ocean afforded by our limited set of fixed stations (Fig. 3).
Our results are affected by substantial decadal oxygen variability (Garcia et al., 2005) caused by large scale changes in atmospheric and oceanic circulation patterns associated with the El Ni ño Southern Oscillation (McPhaden et al., 2006), the North Atlantic Oscillation (Hurrell and Deser, 2009) and the Pacific Decadal Oscillation (Mantua and Hare, 2002) for example.The importance of decadal variality is depicted for a few individual time series on Fig. 2, and is largely responsible for the very different patterns of oxygen trends obtained for different reference periods (Figs. 7 and 8).

Conclusions
In this paper, we have shown that -The median published oxygen trends are more negative than the median trends computed from raw oxygen data, suggesting a publication bias in favor of strongly negative trends that is likely due to the adverse ecosystem implications of hypoxia., 53, 1769-1786, 2006. 9148 Yasuhara, M., Yamazaki, H., Tsujimoto, A., and Hirose, K.: The effect of long-term spatiotemporal variations in urbanization-induced eutrophication on a benthic ecosystem, Osaka Bay, Japan, Limnol.Oceanogr., 52, 1633-1644, 2007. 9148 Yunev, O. A., Carstensen, J., Moncheva, S., Khaliulin, A., Aertebjerg, G., and when they are routinely calibrated against Winkler titrations.But in many laboratories, the lack of regular sensor calibration can lead to inaccurate or suspicious data.The National Oceanographic Data Center (NODC) took the decision to only use oxygen data believed to be obtained by chemical titration methods in their production of the climatological World Ocean Atlas 2005(Garcia et al., 2006), and we followed their example by erring on the side of caution for this global ocean data analysis.With respect to the calculation of trends, only using Winkler titrations offers an additional benefit.It avoids the possibility of introducing artificial jumps in the time series arising from a change in measurement method.We restricted our literature search with the ASFA (Aquatic Sciences and Fisheries Abstracts) and SCOPUS bibliography databases to papers with publication dates of Introduction center and allowed a data search radius of 10 km.Then for each year, we calculated mean oxygen values at the depths of 0 m (0-2 m), 10±1 m, 20±2.5 m, 50±5 m, 75±5 m, 100±5 m, 150± 10 m, 200±10 m, 250±10 m and 300±10 m.To avoid variability in the oxygen time series associated with seasonal cycles, our analyses were limited to the summer months of July, August and September from the sea surface to 75 m depth.At depths of 100 m and greater, where the seasonal cycles are much weaker, we used all available oxygen data from the twelve months of the year.Combining the trend estimates at 0, 50, 100, 150, 200, 250 and 300 m, we also calculated oxygen trend Introduction 98 µmol L −1 yr −1 .The median oxygen trends in the other two distance bands are also negative, but less so than in the coastal band.The story is different for the mean published oxygen trends.The most negative mean trend value (−0.74±0.44 µmol L −1 yr −1 ) is from the open ocean, with mean values that are not different from zero in the other two distance bands.If we consider all cate- . But the time period had the largest effect on DO trends.Between (2009) predicted oxygen concentration trends of −0.04 to −0.07 µmol L −1 yr −1 from 1990-1999 to 2090-2099 in the upper 300 m of the Atlantic and Pacific Oceans.

--
Based on the raw data analysis, oxygen trends in the 1976-2000 period are more negative than in the 1951-1975 period, indicating a recent worsening of hypoxia.Based on the raw data analysis for the 1976-2000 period, oxygen concentrations are declining faster in the coastal ocean (−0.35 ±0.12 µmol L −1 yr −1 ) than in the open ocean (−0.09 ±0.06 µmol L .V., Podymov, O. I., and Yakubenko, V. G.: The northeastern Black Sea redox zone: Hydrochemical structure and its temporal variability, Deep-Sea Res.Pt II

Fig. 5 .
Fig. 5. Histogram of oxygen concentration trends calculated from oxygen time series published in refereed journals (Table2).
also play a role in lowering oxygen concentrations.Figures

2 Oxygen time series from a global dataset
. In some cases, negative oxygen values indicate the presence of H 2 S which Figures

Table 5
).The percentage of oxygen time series with negative trends is greater than 50% at all depths

Table 2 .
List of refereed journal publications from which we obtained oxygen trends, or digitized oxygen time series and then calculated the trends.Oxygen trends are given in µmol L −1 yr −1 and the distances are relative to the nearest shoreline from a continent or island.Introduction

Table 3 .
Oxygen trend statistics (µmol L −1 year −1 ) from published time series in various ranges of distances from the coast.Abbreviations: C.I. = confidence interval; Std Dev = standard deviation; N = number of time series; Perc Neg = Percentage of negative trends.